# João Moura

João Moura is the founder of CrewAI, a company building software for creating and managing AI agents for business work. His posts explain how agents can handle changing goals, coordinate across workflows, and operate with controls such as cost limits and data policies.

- Profile: https://b2binfluencer.marketing/creators/jo-o-moura/
- LinkedIn: https://www.linkedin.com/in/joaomdmoura
- LinkedIn followers: 53,070
- Median of last 8 posts: 121 reactions, 12 comments (checked 2026-11-02)
- Topics: artificial-intelligence, startup, ai-governance, ai-agents, enterprise-ai-adoption
- Based on 8 dated posts, latest 2026-09-29

## Summary

João Moura explains agentic AI for operations leaders, turning agent goals, controls, and enterprise adoption into practical buying decisions.

- Best for: B2B automation or AI platform seeking explainer content for operations and technical leaders.
- Not for: Consumer mass-market or lifestyle campaigns needing high-volume short-form video and broad awareness.
- Check first: Confirm sponsorship offer, rate card, media kit, audience demographics, and prior video performance.

### Takeaways

- An agent is defined by being handed a goal, not a list of steps.
- AI adoption is best described as a ladder of observable workplace behaviors.
- Governance belongs in the execution runtime, not in code or at the network edge.

## Main ideas

- **An agent is defined by being handed a goal, not a list of steps**: Moura draws the line between a workflow and an agent at whether someone wrote out the steps or handed over an outcome. A system following specified steps does the same thing in the same order every day; a system given a goal works out its own path and keeps re-deciding as conditions change. Moura is explicit that this is not an argument against workflows: when the path never changes, fixed steps are cheaper and easier to trust. The reasoning is about where a person's judgment stops being needed, not about which approach is more advanced.
- **AI adoption is best described as a ladder of observable workplace behaviors**: Moura's growth team maps seven stages of using AI at work, each defined by something you can check rather than by capability claims. The stages run from asking a chatbot and copying the answer somewhere yourself, through something that runs before you ask, doing one job on its own, being handed a goal, agents passing work to each other, several systems connecting what a person would otherwise connect mentally, and finally one agent overseeing the rest. Each stage comes with a test question. Moura presents the last stage as a direction builders are heading, not a way companies operate today.
- **Governance belongs in the execution runtime, not in code or at the network edge**: In launching CrewAI's Agent Control Plane, Moura argues that where controls live determines how usable they are. Put guardrails in agent code and every policy update becomes a software release; put them in an external gateway and you get routing and API reuse, but a proxy cannot see inside the agent loop, cap how many reasoning iterations run, pause a run midway or restrict execution to business hours. Because what is being governed is the degree of autonomy delegated to a system, control belongs where that system runs.
- **Enterprise buyers want fewer vendors, even when the market rewards specialists**: Moura tells a story about selling CrewAI to a large bank, where each answer about observability, then memory, pointed to a different product. Moura accepts that the industry is fragmenting into layers, each with companies trying to be the best in the world at one thing, but argues that this pattern looks different from the buyer's side of the table: nobody running a large enterprise wants five contracts, five security reviews or five vendors to chase when something breaks at two in the morning. The resolution Moura describes is building the unglamorous parts anyway, while conceding CrewAI is not best-in-class at each of them. The conclusion drawn is that architecture arguments get settled by procurement rather than by engineers.
- **What enterprises value in integrations is fit with what they already run**: Moura describes building access to thousands of public tool integrations and then learning from customers that they wanted a private registry of the tools their own teams had already built. From outside a platform, more connectors reads as more capability; from inside an enterprise, what matters is what is already wired into your systems, with your authentication around it and your data rules on top. Moura treats that as a mistake in perspective: thinking like someone building a platform instead of someone running an enterprise. A related argument is that standards matter less for access to everything than for preserving the ability to change vendors later without losing the integration work.
- **Model choice and workflow improvement should be measured in separate loops**: Moura frames a recurring engineering question, whether to move crews to a new model, and argues the answer needs measurement rather than a dashboard someone squints at. CrewAI's experimental command-line evaluation tool runs two loops kept deliberately apart. One benchmarks a model against a baseline on quality, cost, speed and tokens, so a frontier model can be used on one role and a cheaper open-weight model on another. The other keeps models fixed and adjusts the prompt, tool definition or routing until a gate clears. Moura flags the guardrail that matters: left unguarded, such a loop will overfit a small evaluation set or rewrite a configuration into a mess purely to turn a failing result into a passing one.
- **Prototypes are easy; production deployment is the hard part**: Moura's position is that many companies can now build an AI prototype, and that the difficulty lies in what comes next: choosing the right business process, integrating it with existing systems, getting it through security and governance, putting it into production and operating it afterwards. This is the gap partnerships are meant to close, and it is also why global expansion, in Moura's telling, means local people, partners and language rather than selling from far away. A related post looks at what remains after an approval, where status is inferred and work moves through email, spreadsheets and follow-ups that can go unnoticed, and argues that the coordination around a decision is where the delay lives.
- **Keep building, stay out of the critical path, and don't mistake recognition for progress**: Alongside the product arguments, Moura writes about how the company is run. Moura still codes on weekends out of enjoyment and as a source of ideas, but stopped taking tasks after learning that working on anything critical makes them a blocker, so ideas go in as pull requests for the team to finish. Improving the product is described as an ongoing effort, including an experiment that made all of CrewAI's agents run five times faster from a small fix that had been missed. The stated aim is that every agent built with CrewAI should be as good by default as a coding agent Moura uses daily. Industry recognition, meanwhile, is treated as validation of the category rather than of the company: a slide does not build a company.

## Work to read

- [Did you hand your AI a list of steps, or a goal?](https://www.linkedin.com/posts/joaomdmoura_did-you-hand-your-ai-a-list-of-steps-or-activity-7509280640908009472-_SkM)
- [How far up the AI ladder are you, really?](https://www.linkedin.com/posts/joaomdmoura_how-far-up-the-ai-ladder-are-you-really-activity-7508556085616611328-pXjW)
- [A while ago I was trying to sell CrewAI to a huge bank](https://www.linkedin.com/posts/joaomdmoura_a-while-ago-i-was-trying-to-sell-crewai-to-activity-7511092545632804864-OuQ6)
- [We built a feature that gave our customers access to 2,000 tool integrations](https://www.linkedin.com/posts/joaomdmoura_we-built-a-feature-that-gave-our-customers-activity-7511454932936388609-OEG9)
- [Your coding agent cannot read your eval dashboard](https://www.linkedin.com/posts/joaomdmoura_your-coding-agent-cannot-read-your-eval-dashboard-activity-7523072744478007296-Kd-s)
- [Today we launch CrewAI's Agent Control Plane](https://www.linkedin.com/posts/joaomdmoura_today-we-launch-crewais-agent-control-plane-activity-7516883869661454336-rn2r)
- [CrewAI is expanding globally, and this one is specially meaningful to me](https://www.linkedin.com/posts/joaomdmoura_crewai-is-expanding-globally-and-this-one-activity-7515419079076245505-W1_-)
- [Someone sent me this photo from a Gartner conference](https://www.linkedin.com/posts/joaomdmoura_someone-sent-me-this-photo-from-a-gartner-activity-7514346714024017920-uJ2p)
- [My engineering team probably wishes I did not code](https://www.linkedin.com/posts/joaomdmoura_my-engineering-team-probably-wishes-i-did-activity-7511756943137513473-j4CD)

## Brand relationships

- **CrewAI** (owned_business)
- **DXC Technology** (disclosed_partnership)
- **Capgemini** (disclosed_partnership)
- **Gartner** (mention)
- **Cursor** (mention)

## Partnerships

No creator sponsorship offer, media kit, rate card, or brand-deal inquiry route appears in the material reviewed for this section. The partnership activity that is documented sits at the company level inside CrewAI's own business: a partnership with DXC Technology to help enterprises across Latin America move agentic AI from pilots into production, described as a company of around 115,000 people operating in more than 70 countries, and a joint live session on 9 November with Arindam Choudhury of Capgemini walking through a single credit decision from approval to funding. Both are CrewAI programs, not placements on Moura's own account, and neither publishes a sponsorship specification, deliverable list, or application process. For a buyer, that means there is no published inventory to brief against: no stated formats, pricing, disclosure terms, or scheduling commitments. The practical reading is that a partnership involving Moura is more likely a business-development conversation tied to CrewAI's product and enterprise deployment work than a standard creator sponsorship. The operational ground the posts cover — agent governance, cost controls, data handling, integration with systems an enterprise already runs — is the context any proposal would need to fit. Whether a sponsorship route exists is not documented here; treat that as a research gap to close before committing budget.

## Alternatives

- Stephen G. Pope: Choose Stephen G. Pope if the campaign should teach teams how to start building a useful AI tool without overengineering it. In a post about an accountability coach, he shows how a rigid chain of steps can become brittle, then explains the agent alternative using memory, tools and a system prompt. Another post recommends starting with one painful task and adding complexity only after it works. Those examples make him a closer fit than João Moura for a hands-on beginner series about workflow design and implementation.
- Othmane Khadri: Choose Othmane Khadri if the same campaign should center on practical sales and go-to-market workflows. One post describes using an agent to review lost CRM deals, group objections and draft follow-ups; another describes an agent that responds to unwanted AI-generated comments. These concrete sales and outreach examples could make the campaign more relevant to revenue teams evaluating agentic workflows in their own work.
- Max Thilén: Choose Max Thilén if the buyer task is specifically to explain how a marketing team can work with AI. His post lays out operating practices including shared business context, connecting agents to marketing systems, measuring quality, setting safety boundaries and keeping human checks. It also gives concrete examples, such as building a targeted account list from conference attendees and finding usable clips in a podcast archive.

---
AI-assisted summary of public posts. Audience demographics are not available. Engagement appears only where we checked recent posts. Method: https://b2binfluencer.marketing/methodology/